To identify novel targets for metabolic engineering of riboflavin production, we generated about 10,000 random, transposon-tagged mutants of an industrial, riboflavin-producing strain of Bacillus subtilis. Process-relevant screening conditions were established by developing a 96-deep-well plate method with raffinose as the carbon source, which mimics, to some extent, carbon limitation in fed batch cultures. Screening in raffinose and complex LB medium identified more efficiently riboflavin overproducing and underproducing mutants, respectively. As expected for a "loss of function" analysis, most identified mutants were underproducers. Insertion mutants in two genes with yet unknown function, however, were found to attain significantly improved riboflavin titers and yields. These genes and possibly further ones that are related to them are promising candidates for metabolic engineering. While causal links to riboflavin production were not obvious for most underproducers, we demonstrated for the gluconeogenic glyceraldehyde-3-phosphate dehydrogenase GapB how a novel, non-obvious metabolic engineering strategy can be derived from such underproduction mutations. Specifically, we improved riboflavin production on various substrates significantly by deregulating expression of the gluconeogenic genes gapB and pckA through knockout of their genetic repressor CcpN. This improvement was also verified under the more process-relevant conditions of a glucose-limited fed-batch culture.